Most AI chat tools reset every session. You start fresh, re-explain yourself, and the conversation never builds into anything. HLT's memory system works differently, and understanding how it works helps you get more out of it.
The problem with stateless chat ¶
A stateless chatbot treats every conversation as the first one. It has no record of what you said yesterday, what topics matter to you, or how the last exchange ended. That is fine for customer support, where the goal is to close a ticket. It is not fine for a companion, where the goal is continuity. The feeling of being known by someone requires that they actually remember you.
What a memory graph does differently ¶
HLT uses a persistent memory graph rather than a simple conversation log. The graph tags key details from each session: names you mention, topics that come up repeatedly, emotional tone at the end of a conversation, and specific events you describe. These tags are weighted by recency and frequency, so the character does not surface irrelevant details from six months ago but does remember what you talked about last week.
How emotional context is tracked ¶
Since the May 2026 update, the memory system also tracks emotional context at the session level. If a conversation ends on a heavy or unresolved note, the character's opening in the next session reflects that. It will not ignore the gap or launch into cheerful small talk. This behavior was rebuilt from scratch over six weeks because the original tagging model only worked on the most recent session, not across the full history window.
What the different plan tiers retain ¶
Free plan users have a 7-day history window. Premium users have 90 days. Ultimate users have 365 days. The memory graph is active on all plans, but the depth of what the character can draw on depends on how much history is retained. If you are on Free and want to preserve a thread, upgrading before the 7-day window closes will carry the full history forward.
Practical tips for better memory ¶
The memory system works best when you are specific. Mentioning your dog's name, a project you are working on, or a place you visited gives the graph concrete tags to work with. Vague statements like 'I had a hard day' are logged but carry less weight than 'the presentation at work went badly.' The more specific you are, the more the character has to work with in future sessions.
Memory is not a feature we added late. It was the first design decision Marcus Hale wrote on the whiteboard in 2022. If you want to see it in action, the Free plan is a reasonable starting point.